Research

I am a PhD student at the Centre for Doctoral Training in Computational Statistics and Data Science at the University of Bristol, supervised by Professor Patrick Rubin-Delanchy and Professor Nick Whiteley.

My research is involved with statistical inference on graphs. In particular, applying spectral methods to graph embeddings, and harnessing the low-dimensional manifold structure that can arise from these models.

I recently completed a 6-month internship at Amazon, in which I researched the problem of hallucination detection in LLM responses. As well as implementing various LLM evaluation methods to be used in conjunction with the teams LLM applications, I developed a novel technique GraphEval, based on representing information in Knowledge Graph structures.

Publications

How high is ‘high’? Rethinking the roles of dimensionality in topological data analysis and manifold learning.

Hannah Sansford, Nick Whiteley, Patrick Rubin-Delanchy, (2025).

(Under Review)

GraphEval: A Knowledge-Graph Based LLM Hallucination Evaluation Framework.

Hannah Sansford, Nicholas Richardson, Hermina Petric Maretic, Juba Nait Saada, (2024).

KiL'24: Workshop on Knowledge-infused Learning co-located with 30th ACM KDD Conference.

Selected for oral presentation.

Implications of sparsity and high triangle density for graph representation learning.

Hannah Sansford, Alexander Modell, Nick Whiteley, Patrick Rubin-Delanchy, (2023).

International Conference on Artificial Intelligence and Statistics (AISTATS).

Selected for oral presentation (<2% of all submissions).

Recent and Upcoming Talks

KiL'24: Workshop on Knowledge-infused Learning co-located with 30th ACM KDD Conference, August 26, 2024, Barcelona, Spain (Paper presentation).
Harvard University, May 2023 (Invited online seminar).
International Conference on Artificial Intelligence and Statistics (AISTATS), Valencia, Spain, April 2023 (Paper presentation).
Compass Annual Conference, Bristol, U.K., September 2022 (Invited talk).

Page updated

Google Sites

Report abuse